CN110180186A - A kind of topographic map conversion method and system - Google Patents
A kind of topographic map conversion method and system Download PDFInfo
- Publication number
- CN110180186A CN110180186A CN201910453305.0A CN201910453305A CN110180186A CN 110180186 A CN110180186 A CN 110180186A CN 201910453305 A CN201910453305 A CN 201910453305A CN 110180186 A CN110180186 A CN 110180186A
- Authority
- CN
- China
- Prior art keywords
- geometric
- label
- picture
- matched
- topographic map
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 70
- 238000006243 chemical reaction Methods 0.000 title claims abstract description 38
- 238000013528 artificial neural network Methods 0.000 claims abstract description 68
- 238000007781 pre-processing Methods 0.000 claims abstract description 9
- 238000011156 evaluation Methods 0.000 claims description 27
- 238000013527 convolutional neural network Methods 0.000 claims description 25
- 239000013598 vector Substances 0.000 claims description 25
- 230000002457 bidirectional effect Effects 0.000 claims description 23
- 230000002452 interceptive effect Effects 0.000 claims description 22
- 230000008569 process Effects 0.000 claims description 22
- 238000012545 processing Methods 0.000 claims description 22
- 230000015654 memory Effects 0.000 claims description 17
- 238000004422 calculation algorithm Methods 0.000 claims description 13
- 238000012512 characterization method Methods 0.000 claims description 13
- 230000000694 effects Effects 0.000 claims description 13
- 239000011159 matrix material Substances 0.000 claims description 13
- 230000001537 neural effect Effects 0.000 claims description 10
- 238000000605 extraction Methods 0.000 claims description 9
- 238000005260 corrosion Methods 0.000 claims description 7
- 230000007797 corrosion Effects 0.000 claims description 7
- 238000005516 engineering process Methods 0.000 abstract description 3
- 238000012549 training Methods 0.000 description 22
- 238000013480 data collection Methods 0.000 description 19
- 238000010586 diagram Methods 0.000 description 14
- 230000006403 short-term memory Effects 0.000 description 9
- 238000001514 detection method Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 7
- 230000003628 erosive effect Effects 0.000 description 4
- 238000003860 storage Methods 0.000 description 4
- 238000007792 addition Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000005520 cutting process Methods 0.000 description 3
- 230000000306 recurrent effect Effects 0.000 description 3
- 230000010339 dilation Effects 0.000 description 2
- 238000012015 optical character recognition Methods 0.000 description 2
- 238000002360 preparation method Methods 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 241001270131 Agaricus moelleri Species 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000005530 etching Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000007670 refining Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Classifications
-
- A—HUMAN NECESSITIES
- A63—SPORTS; GAMES; AMUSEMENTS
- A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
- A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
- A63F13/50—Controlling the output signals based on the game progress
- A63F13/53—Controlling the output signals based on the game progress involving additional visual information provided to the game scene, e.g. by overlay to simulate a head-up display [HUD] or displaying a laser sight in a shooting game
- A63F13/537—Controlling the output signals based on the game progress involving additional visual information provided to the game scene, e.g. by overlay to simulate a head-up display [HUD] or displaying a laser sight in a shooting game using indicators, e.g. showing the condition of a game character on screen
- A63F13/5378—Controlling the output signals based on the game progress involving additional visual information provided to the game scene, e.g. by overlay to simulate a head-up display [HUD] or displaying a laser sight in a shooting game using indicators, e.g. showing the condition of a game character on screen for displaying an additional top view, e.g. radar screens or maps
-
- A—HUMAN NECESSITIES
- A63—SPORTS; GAMES; AMUSEMENTS
- A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
- A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
- A63F13/60—Generating or modifying game content before or while executing the game program, e.g. authoring tools specially adapted for game development or game-integrated level editor
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Optics & Photonics (AREA)
- Radar, Positioning & Navigation (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of topographic map conversion method and systems, are related to image-editing technology field, solve the problems, such as that there are inefficiency, cumbersome when player edits topographic map in the prior art.This method comprises: pre-processing to rough draft topographic map, interference picture is removed in acquisition；By deep neural network, identification goes in interference picture to indicate the label of landform element type and geometric parameter；For the label that will identify that from going in interference picture to delete, label picture is removed in acquisition；Identification goes in label picture to indicate the geometric figure of landform element outline；The label that will identify that is matched with the geometric figure identified；According to the position of the matched label of geometric figure, geometric figure and geometric figure that identify, rough draft topographic map is converted into editable electronic topographic map.The rough draft topographic map of player can be converted into editable electronics blueprint in game editing machine automatically by the solution of the present invention, implementation is simple, conveniently, it is easy to operate, and improve treatment effeciency.
Description
Technical field
The present invention relates to image-editing technology field more particularly to a kind of topographic map conversion methods and system.
Background technique
With the continuous development of electronic technology, online game has become the mode of many consumer entertainment pastimes, network trip
The type and content of play are also more and more abundant, while user is also higher and higher to the experience requirements of online game.In order to meet not
With the demand for experience of user, player can designed, designed topographic map, but in traditional game map_editor, player edits topographic map
When there is a problem of inefficiency, cumbersome, this problems demand solves.
Summary of the invention
The embodiment of the present invention provides a kind of topographic map conversion method and system, edits landform to solve player in the prior art
There is a problem of inefficiency, cumbersome when figure.
In order to solve the above-mentioned technical problem, the present invention is implemented as follows:
In a first aspect, the embodiment provides a kind of topographic map conversion methods, comprising:
Rough draft topographic map is pre-processed, interference picture is removed in acquisition；
By deep neural network, go in interference picture to indicate the mark of landform element type and geometric parameter described in identification
Label；
The label that will identify that goes in interference picture to delete from described, and label picture is removed in acquisition；
It goes in label picture to indicate the geometric figure of landform element outline described in identification；
The label that will identify that is matched with the geometric figure identified；
According to the position of the matched label of geometric figure, geometric figure and geometric figure that identify, by rough draft topographic map
It is converted into editable electronic topographic map.
Optionally, described to pre-process to rough draft topographic map, obtaining the step of removing interference picture includes:
Binary conversion treatment is carried out to rough draft topographic map, obtains binaryzation picture；
Image noise and interfering line to binaryzation picture are removed processing, and interference picture is removed in acquisition.
Optionally, described by deep neural network, remove in interference picture to indicate landform element type and several described in identification
The step of label of what parameter includes:
By the first deep neural network, identifies and intercept and described go in interference picture to indicate landform element type and geometry
The label preselected area of parameter；
After the range of label preselected area is expanded by parameter preset, the interfering line in label preselected area is removed
Processing, and the coordinate position of label preselected area is adjusted, obtain the label picture for being marked with the preparatory region of label；
By the second deep neural network, identifying indicates landform element class in each label preselected area in the label picture
The icon of type and the text for indicating landform element geometric parameter.
Optionally, first deep neural network is long including the first deep layer convolutional neural networks layer, the first bidirectional circulating
Short-term memory neural net layer, the first full articulamentum and the first output layer；
It is described by the first deep neural network, identify and intercept it is described go in interference picture to indicate landform element type and
The step of label preselected area of geometric parameter includes:
It goes interference picture as the input of the first deep layer convolutional neural networks layer for described, passes through the first deep layer convolution
Neural net layer goes interference picture to carry out feature extraction to described, obtains first network characteristic pattern；
The first network characteristic pattern is slided, every row obtains W with the window of predetermined size to slide unit by turn line by line
A one-dimensional vector, W are the width of the first network characteristic pattern；
Using the W one-dimensional vector that every row obtains as the defeated of the long Memory Neural Networks layer in short-term of the first bidirectional circulating
Enter, the first tensor is obtained by the long Memory Neural Networks layer in short-term of first bidirectional circulating；
Using first tensor as the input of the first full articulamentum, second is obtained by the described first full articulamentum
Amount；
Using second tensor as the input of the first output layer, the first network is obtained by first output layer
Each pixel mapped goes the output of pixel region in interference picture as a result, the output result includes vertical sits in characteristic pattern
Mark offset prediction result, text probabilistic forecasting result and boundary shifts amount prediction result；
According to pixel mapped each in the network characterization figure go interference picture in pixel region output as a result,
It goes in interference picture to indicate the label preselected area of landform element type and geometric parameter described in determination.
Optionally, described using second tensor as the input of the first output layer, it is obtained by first output layer
Each pixel mapped removes the step of output result of pixel region in interference picture packet in the first network characteristic pattern
It includes:
Using second tensor as the input of the first output layer, the first network is obtained by first output layer
Each pixel mapped under different anchor anchor sizes goes the output result of pixel region in interference picture in characteristic pattern.
Optionally, second deep neural network is long including the second deep layer convolutional neural networks layer, the second bidirectional circulating
Short-term memory neural net layer, the second full articulamentum and the second output layer；
It is described by the second deep neural network, identifying indicates that landform is first in each label preselected area in the label picture
The icon of plain type and the step of indicating the text of landform element geometric parameter include:
Using the label picture as the input of the second deep layer convolutional neural networks layer, pass through the second deep layer convolution mind
Feature extraction is carried out to label preselected area each in the label picture through network layer, obtains the second network characterization figure；
Using the second network characterization figure as the input of the long Memory Neural Networks layer in short-term of the second bidirectional circulating, pass through institute
It states the long Memory Neural Networks layer in short-term of the second bidirectional circulating and obtains third tensor；
Using the third tensor as the input of the second full articulamentum, the 4th is obtained by the described second full articulamentum
Amount；
Using the 4th tensor as the input of the second output layer, it is special that two network is obtained by second output layer
Every frame result vector is mapped as the probability of all icons and text in sign figure；
Be mapped as the probability of all icons and text according to frame result vector every in the two network characterizations figure, determine described in
The text for marking the icon for indicating landform element type in picture in each label preselected area and indicating landform element geometric parameter.
Optionally, described to pass through the first deep neural network, it identifies and intercepts and described go in interference picture to indicate landform member
Before the label preselected area of plain type and geometric parameter, further includes:
By pre-prepd first rough draft label data collection, the first deep neural network is trained, and in training
Increase interfering line and interference text in the process；
Wherein, the first rough draft label data collection includes icon data set and lteral data collection；
It is described by the second deep neural network, identifying indicates that landform is first in each label preselected area in the label picture
Before the icon of plain type and the text of expression landform element geometric parameter, further includes:
By pre-prepd second rough draft label data collection, the second deep neural network is trained, and in training
Increase interfering line and interference text in the process；
Wherein, the second rough draft label data collection includes icon data set and lteral data collection.
Optionally, gone described in the identification in label picture indicate landform element outline geometric figure the step of include:
Label picture is gone to carry out an image thinning according to default thinning algorithm by described, and treated to image thinning
It goes label picture to carry out expansion process and corrosion treatment, obtains first stage picture；
Label picture will be gone to carry out logical AND operation before the first stage picture and image thinning processing, obtains second
Stage picture；
The second stage picture is subjected to secondary image refinement according to default thinning algorithm, obtains phase III picture；
The profile point for searching geometric figure contour line by way of seeking all in the phase III picture, according to seeking all over
As a result the geometric figure that landform element outline is indicated in the phase III picture is determined.
Optionally, the profile point of geometric figure contour line is searched in the way of seeking all in the phase III picture,
It is determined in the phase III picture after the geometric figure of expression landform element outline according to result is sought all over, further includes:
The evaluation parameter for obtaining each geometric figure determined in the phase III picture, by the evaluation of each geometric figure
Parameter is compared with preset evaluation index respectively；
Delete the geometric figure that evaluation parameter in the phase III picture is less than preset evaluation index；
Wherein, the evaluation parameter includes profile minimum image vegetarian refreshments number and profile minimum circumscribed rectangle area.
Optionally, the step of label that will identify that is matched with the geometric figure identified include:
The label that will identify that is as target labels, according to the relative positional relationship of target labels and each geometric figure,
Determine the geometric figure to be matched of target labels；
If geometric figure to be matched, which does not match, gives other labels, using the geometric figure to be matched as of target labels
With geometric figure；
If geometric figure to be matched, which has matched, gives other labels, obtain target labels minimum circumscribed rectangle with it is to be matched several
What intersection of figure minimum circumscribed rectangle and the first ratio of union, and obtain matched label minimum circumscribed rectangle with it is to be matched
The intersection of geometric figure minimum circumscribed rectangle and the second ratio of union；
If the first ratio is greater than the second ratio, using geometric figure to be matched as the matching geometric figure of target labels；
If the second ratio is greater than the first ratio, geometric figure to be matched is continued as the matching geometric figure for having matched label；
If the first ratio and the second ratio are identical, target labels minimum circumscribed rectangle central point and geometry to be matched are obtained
The first distance of figure minimum circumscribed rectangle central point, and obtain matched label minimum circumscribed rectangle central point with it is to be matched several
The second distance of what figure minimum circumscribed rectangle central point；
If first distance is less than second distance, using geometric figure to be matched as the matching geometric figure of target labels；
If second distance is greater than first distance, geometric figure to be matched is continued as the matching geometric figure for having matched label.
Optionally, the relative positional relationship according to target labels and each geometric figure, determine target labels to
Match geometric figure the step of include:
Searching overlay area and target labels has the geometric figure of intersection；
If the overlay area of only one geometric figure and the overlay area of target labels have intersection, by the geometric graph
To be matched geometric figure of the shape as target labels；
If having the overlay area of multiple geometric figures and the overlay area of target labels that there is intersection, there is friendship by multiple
The geometric figure of collection obtains target labels minimum circumscribed rectangle and each candidate figure minimum circumscribed rectangle as candidate figure
The ratio of intersection and union, and obtain to be matched geometric figure of the maximum candidate figure of ratio as target labels；
If being not covered with region and target labels has the geometric figure of intersection, target labels minimum circumscribed rectangle is obtained
Central point is obtained apart from the smallest geometric figure at a distance from each geometric figure minimum circumscribed rectangle central point as target
The geometric figure to be matched of label.
Optionally, the position of the basis identifies geometric figure, the matched label of geometric figure and geometric figure, will
Rough draft topographic map is converted into the step of editable electronic topographic map and includes:
According to the position of the matched label of geometric figure, geometric figure and geometric figure that identify, rough draft is determined respectively
Profile, type, geometric parameter and the position of topographic map mesorelief element；
According to the type and geometric parameter of landform element, the grayscale image of performance landform element stereoscopic effect is obtained；
Profile and grayscale image to landform element carry out distortion processing respectively, according to the profile of the landform element after distortion and
Grayscale image obtain performance landform element stereoscopic effect, and with the matched height map of landform element shape；
According to the position of landform element, the 3-dimensional digital matrix that the height map of all landform elements is added to pre-generated
In；
According to superimposed 3-dimensional digital matrix, editable electronic topographic map is generated.
Second aspect, the embodiments of the present invention also provide a kind of topographic map converting systems, comprising:
Preprocessing module, for pre-processing to rough draft topographic map, interference picture is removed in acquisition；
First identification module goes in interference picture to indicate landform element class described in identification for passing through deep neural network
The label of type and geometric parameter；
Removing module, the label for will identify that go in interference picture to delete from described, and label picture is removed in acquisition；
Second identification module, for identification geometric figure for going in label picture to indicate landform element outline；
Matching module, the label for will identify that are matched with the geometric figure identified；
Conversion module, for according to the position of the matched label of geometric figure, geometric figure and geometric figure identified,
Rough draft topographic map is converted into editable electronic topographic map.
Optionally, first identification module includes:
First identification submodule described removes table in interference picture for by the first deep neural network, identifying and intercepting
Show the label preselected area of landform element type and geometric parameter；
Submodule is removed, after expanding the range of label preselected area by parameter preset, in label preselected area
Interfering line be removed processing, and adjust the coordinate position of label preselected area, obtain the mark for being marked with the preparatory region of label
Remember picture；
Second identification submodule, for by the second deep neural network, identifying each label pre-selection in the label picture
The icon of landform element type is indicated in region and indicates the text of landform element geometric parameter.
In embodiments of the present invention, rough draft topographic map is pre-processed first, interference picture is removed in acquisition；Then pass through depth
Neural network is spent, identification goes in interference picture to indicate the label of landform element type and geometric parameter；The label that will identify that again
From going in interference picture to delete, after label picture is removed in acquisition, identification goes in label picture to indicate the geometric graph of landform element outline
Shape；Then the label that will identify that is matched with the geometric figure identified；Geometric figure, the geometry that last basis identifies
Rough draft topographic map is converted into editable electronic topographic map by the label of Graphic Pattern Matching and the position of geometric figure.So pass through this
The rough draft topographic map of player can be converted into editable electronics blueprint in game editing machine automatically by the method for inventive embodiments,
Player need to only determine the geomorphic type in specified region in rough draft topographic map by icon, describe landforms by geometric figure lines
Chamfered shape, without executing other operations, implementation is simple, conveniently, it is easy to operate, and improve treatment effeciency.
Detailed description of the invention
Fig. 1 is the flow chart of topographic map conversion method provided in an embodiment of the present invention；
Fig. 2 is the picture schematic diagram in topographic map conversion method provided in an embodiment of the present invention Jing Guo binary conversion treatment；
Fig. 3 is the picture signal removed in topographic map conversion method provided in an embodiment of the present invention by noise and interfering line
Figure；
Fig. 4 is the picture schematic diagram for expanding label preselected area in topographic map conversion method provided in an embodiment of the present invention；
Fig. 5 is the signal that interfering line in label preselected area is removed in topographic map conversion method provided in an embodiment of the present invention
Figure；
Fig. 6 is the picture signal of precise marking label preselected area in topographic map conversion method provided in an embodiment of the present invention
Figure；
Fig. 7 is to remove label picture schematic diagram in topographic map conversion method provided in an embodiment of the present invention；
Fig. 8 is to increase interfering line schematic diagram one in topographic map conversion method provided in an embodiment of the present invention；
Fig. 9 is to increase interfering line schematic diagram two in topographic map conversion method provided in an embodiment of the present invention；
Figure 10 is the result schematic diagram in topographic map conversion method provided in an embodiment of the present invention after dilation erosion；
Figure 11 is the result schematic diagram of dilation erosion after refining in topographic map conversion method provided in an embodiment of the present invention；
Figure 12 is the result schematic diagram in topographic map conversion method provided in an embodiment of the present invention after secondary refinement；
Figure 13 is the schematic diagram one that profile point is traversed in topographic map conversion method provided in an embodiment of the present invention；
Figure 14 is the schematic diagram two that profile point is traversed in topographic map conversion method provided in an embodiment of the present invention；
Figure 15 is the schematic diagram that profile is distorted in topographic map conversion method provided in an embodiment of the present invention；
Figure 16 is the schematic diagram that concave polygon is distorted in topographic map conversion method provided in an embodiment of the present invention；
Figure 17 is the schematic diagram that grayscale image is distorted in topographic map conversion method provided in an embodiment of the present invention；
Figure 18 is the structural schematic diagram of topographic map converting system provided in an embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are some of the embodiments of the present invention, instead of all the embodiments.Based on this hair
Embodiment in bright, every other implementation obtained by those of ordinary skill in the art without making creative efforts
Example, shall fall within the protection scope of the present invention.
In some embodiments of the invention, a kind of topographic map conversion method is provided, shown referring to Fig.1, the method
Include:
Step 101: rough draft topographic map being pre-processed, interference picture is removed in acquisition.
Without limitation to the acquisition pattern of rough draft topographic map, such as player can pass through Freehandhand-drawing under line to the embodiment of the present invention
Mode completes rough draft topographic map, then can be by mobile phone, computer etc. eventually by taking pictures or scanning the system of uploading to or player
The mapping software at end, which draws, completes rough draft topographic map, then by picture transfer to system.
But usual player draws and the rough draft topographic map uploaded is RGB triple channel picture, be limited to different paper textures,
Illumination condition and shooting background etc. can generate many redundancies, interfere subsequent identification judgement, in order to promote recognition effect,
For this step by pre-processing to rough draft topographic map, interference picture is clearly removed in acquisition, can accurately identify figure convenient for subsequent
Label and geometric figure in piece.
Step 102: by deep neural network, going in interference picture to indicate landform element type and geometric parameters described in identification
Several labels.
Here, it by deep neural network, accurately can go in interference picture to identify outgoing label by pretreated, with
Determine the type and geometric parameter of rough draft topographic map mesorelief element.
Wherein, the type of landform element such as may include high mountain, river, Hu Bo, grassland, sandy beach, road, but be not limited to
This.
Wherein, the geometric parameter of landform element may include the basic parameter to form landform element, specifically, landform element
Geometric parameter can correspond to the type setting of landform element, if the type of landform element is high mountain, the then geometry of landform element
Parameter may include the relative altitude etc. of high mountain, if the type of landform element is road, then the geometric parameter of landform element can wrap
Include the length and width etc. of road.Certainly, the geometric parameter of above-mentioned landform element is only for example, the embodiment of the present invention
The geometric parameter of landform element is not limited to above-mentioned several.
Step 103: the label that will identify that goes in interference picture to delete from described, and label picture is removed in acquisition.
Here, by the label that will identify that from going in interference picture to delete, label picture is removed in acquisition, convenient for subsequent from figure
Geometric figure is identified in piece.
Step 104: going in label picture to indicate the geometric figure of landform element outline described in identification.
Here, by identifying the geometric figure gone in label picture, it is capable of determining that rough draft topographic map mesorelief element
Chamfered shape.
Step 105: the label that will identify that is matched with the geometric figure identified.
Here, the geometric figure in rough draft topographic map and label are individually drawn, and pass through the label that will identify that and knowledge
Not Chu geometric figure matched, be capable of determining that the type of the Contour matching of rough draft topographic map mesorelief element and several
What parameter, is accurately transformed into electronic topographic map convenient for the subsequent landform element by rough draft topographic map.
Step 106:, will be careless according to the position of the matched label of geometric figure, geometric figure and geometric figure that identify
Original text topographic map is converted into editable electronic topographic map.
Here, it according to the position of the matched label of geometric figure, geometric figure and geometric figure that identify, can determine
The position of the chamfered shape of rough draft topographic map mesorelief element, the type of landform Match of elemental composition and geometric parameter and landform element out
It sets, so that rough draft topographic map is converted into editable electronic topographic map.
The rough draft topographic map of player can be converted into game editor automatically by the topographic map conversion method of the embodiment of the present invention
Editable electronics blueprint in device, player need to only be determined the geomorphic type in specified region by icon in rough draft topographic map, led to
Cross geometric figure lines and describe landforms chamfered shape, without executing other operations, implementation is simple, conveniently, it is easy to operate, and mention
High treatment effeciency.
Optionally, above-mentioned steps 101 include:
Step 1011: binary conversion treatment being carried out to rough draft topographic map, obtains binaryzation picture.
Here, adaptive threshold binary conversion treatment is carried out to rough draft topographic map first, generates binaryzation picture, at this time as schemed
Shown in 2, most of texture and interference are had been removed in rough draft topographic map.
Step 1012: image noise and interfering line to binaryzation picture are removed processing, and interference picture is removed in acquisition.
Here, continue in binaryzation picture due to shooting environmental etc. caused by image noise and interfering line go
Except processing, at this time as shown in figure 3, obtaining clearly binaryzation picture.
Wherein, the image noise and interfering line of binaryzation picture can be eliminated by opening operation (first corrode and expand afterwards), but not
It is limited to this.
Optionally, above-mentioned steps 102 include:
Step 1021: by the first deep neural network, identifying and intercept and described go in interference picture to indicate landform element
The label preselected area of type and geometric parameter.
Here, it goes in interference picture to identify and intercept out by pretreated by the first deep neural network first
Label preselected area further accurately identifies the position of label convenient for subsequent step.
Step 1022: after the range of label preselected area is expanded by parameter preset, to the interference in label preselected area
Line is removed processing, and adjusts the coordinate position of label preselected area, obtains the label picture for being marked with the preparatory region of label.
Since the range of label preselected area is not necessarily very accurate, cannot sometimes completely include icon to be predicted and
Text information needs its position accurately to be adjusted and interfered removal.In this step, to guarantee whole icons and text letter
Breath is all selected, first expands the range of label preselected area by parameter preset, and whole icons and text are all by label at this time
Preselected area includes.But the rear portion geometric figure line segment of expansion be also possible to can by comprising in label preselected area,
As shown in figure 4, can be to identifying that icon and text interfere in process in next step.This step is further by label pre-selected zone
Interfering line in domain is removed processing, as shown in figure 5, the simultaneously coordinate position of intense adjustment label preselected area, is only wrapped
Label preselected area containing icon and text information, at this time can be with the location of pixels of precise marking all icons and text, such as Fig. 6
It is shown, the convenient identification to icon and text, and can be used for it is subsequent icon and text are accurately deleted from original image, label is removed in acquisition
Picture, as shown in fig. 7, to the convenient identification to geometric figure.
Wherein, by the range of label preselected area by parameter preset expand when, such as can by label preselected area laterally and
It is longitudinal to expand 40 pixels respectively, but not limited to this.
Step 1023: by the second deep neural network, identifying in the label picture is indicated in each label preselected area
The icon of landform element type and the text for indicating landform element geometric parameter.
Here, by the second deep neural network, can be recognized accurately in each label preselected area indicates landform element
The icon of type and the text for indicating landform element geometric parameter, so that it is determined that the type and geometric parameter of landform element.
In recent years, the continuous advancement with deep learning algorithm and breakthrough, computer vision object detection task achieve
Huge progress, subset of the text target detection as object detection task, due to the application market of its magnanimity, such as OCR
Multilingual translation etc., is also cured under (Optical Character Recognition, optical character identification) task, natural scene
Hair is taken seriously.
It is different from traditional natural target detection, the target text sequence of text target detection usually has random length, boundary
The mature target detection modelling effect of the characteristics such as fuzzy, directly application is unsatisfactory, therefore uses in the embodiment of the present invention by text
Sequence is height by width, text actual height of 16 pixels, is small scale character block by original character cutting, and neural network is only
The score of the vertical each small scale character block of prediction, is finally spliced into long word sequence.Further, since the reason cut by force, often
Text information that a character block includes is simultaneously imperfect, needs to combine its lateral context auxiliary judgment, therefore in traditional CNN
After (Convolutional Neural Networks, convolutional neural networks), it is complete to joined bidirectional circulating recurrent neural network
It is apt to lateral contextual information, increases the robustness of model.It introduces in detail below.
Optionally, first deep neural network is long including the first deep layer convolutional neural networks layer, the first bidirectional circulating
Short-term memory neural net layer, the first full articulamentum and the first output layer.
In the embodiment of the present invention, the first deep neural network used is by the first deep layer convolutional neural networks layer, first pair
To circulation long short-term memory (B-LSTM, Bidirectional-Long Short Term Memory) neural net layer, first
Full articulamentum and the first output layer are constituted.
Above-mentioned steps 1021 include:
Step 10211: going interference picture as the input of the first deep layer convolutional neural networks layer for described, pass through described the
One deep layer convolutional neural networks layer goes interference picture to carry out feature extraction to described, obtains first network characteristic pattern.
Here, interference picture is gone to initially enter the first deep layer convolutional neural networks layer extraction picture spy by pretreated
Sign obtains first network characteristic pattern.The common VGG16 network architecture such as can be used in first deep layer convolutional neural networks layer, passes through
Final first network characteristic pattern is obtained after five groups of convolution pond layers, having a size of 512 (the 16 of original image chip size Wx H x
/ mono-).
Step 10212: the first network characteristic pattern is slided with the window of predetermined size to slide unit by turn line by line
Dynamic, every row obtains W one-dimensional vector, and W is the width of the first network characteristic pattern.
It here, is to slide with the window of predetermined size by the first network characteristic pattern of the first deep layer convolutional neural networks layer output
Dynamic unit, is slided, every row obtains W one-dimensional vector by turn line by line.Such as by first network characteristic pattern with the group of windows of 3x3x512
At one-dimensional vector as sliding unit, slides laterally by turn line by line and (slide to be mapped on input picture every time and has slided 16 pictures
Element, this sliding distance are determined by 4 maximum value pond layers in convolutional network structure), every row obtains W (characteristic pattern width)
A dimension is the one-dimensional vector of 3x 3x C, and C is port number.
Step 10213: using the W one-dimensional vector that every row obtains as the long short-term memory nerve net of the first bidirectional circulating
The input of network layers obtains the first tensor by the long Memory Neural Networks layer in short-term of first bidirectional circulating.
Here, Memory Neural Networks B- in short-term is grown using the W one-dimensional vector that every row obtains as the first bidirectional circulating
LSTM layers of input obtains the first tensor.For example, using the one-dimensional vector that W dimension of a line is 3x 3x C as B-LSTM layers
Input obtains the W 256 hidden state vector of dimension (being spliced by the 128 hidden state vectors of dimension of 2 unidirectional LSTM networks), to institute
There is row to carry out aforesaid operations, obtains the first tensor of W x H x 256.
Step 10214: using first tensor as the input of the first full articulamentum, being obtained by the described first full articulamentum
Obtain the second tensor.
Here, the first tensor of W x H x 256 is such as inputted into the first full articulamentum, obtains the tensor of W x H x 512.
Step 10215: using second tensor as the input of the first output layer, institute being obtained by first output layer
It states each pixel mapped in first network characteristic pattern and goes the output of pixel region in interference picture as a result, the output is tied
Fruit includes ordinate offset prediction result, text probabilistic forecasting result and boundary shifts amount prediction result.
Here, i.e. the second tensor of result before is inputted into the first output layer, each pixel in first network characteristic pattern
Point obtains three groups of outputs, is each responsible for ordinate offset prediction result, text probabilistic forecasting result and the prediction of boundary shifts amount
As a result.Since (practical receptive field is the 228x228 pixel region in original image to each pixel in first network characteristic pattern
Domain) be mapped to after interference picture between be divided into 16 pixels, therefore each pixel in first network characteristic pattern be responsible for it is pre-
Surveying and removing fixed width in interference picture is the information of 16 pixel regions, comprising: the region is the probability of character block, the region text
Height and vertical centre the point position of block, the boundary shifts amount of the region character block.
Step 10216: pixel region in interference picture is gone according to pixel mapped each in the network characterization figure
Output as a result, determine described in go interference picture in indicate landform element type and geometric parameter label preselected area.
At this point, the final result exported by the first deep neural network, can determine away in interference figure piece indicates ground
The label preselected area of shape element type and geometric parameter.
Optionally, above-mentioned steps 10215 include:
Using second tensor as the input of the first output layer, the first network is obtained by first output layer
Each pixel mapped under different anchor anchor sizes goes the output result of pixel region in interference picture in characteristic pattern.
Here, it in order to meet the multi-scale prediction characteristic of model, increases common in target detection neural network
Anchor mechanism, since character block width has been fixed as 16 pixels, anchor different at this time only is responsible for distinguishing in height side
The character block of face different scale such as can define 9 anchor scales, respectively represent the high pre- measurement ruler of 11 pixels up to 273 pixels
Degree, when making training set, only scale and the immediate anchor of character block just can be positive example, remaining example that is negative by table.
It is as shown in the table by the output result of the first output layer:
Wherein, logarithm loss function can be used for first item prediction result, for second and third loss function, can adopts
With MSE loss function.
Optionally, before above-mentioned steps 1021, further includes:
By pre-prepd first rough draft label data collection, the first deep neural network is trained, and in training
Increase interfering line and interference text in the process；Wherein, the first rough draft label data collection includes icon data set and text number
According to collection.
Here, since deep neural network needs mass data to be trained, and satisfactory training set is obtained very
Difficulty, therefore the first deep neural network of this step is in the training process, and training set is constructed by the way of manually generated, that is, is passed through
Pre-prepd first rough draft label data collection is trained.
Specifically, can be from disclosed MNIST handwritten numeral data set and several hand-written icons of our preparations, at random
Icon, digital jointing are selected as character string, the number spliced is scaled at random high -65 pixels of 16 pixels.It
The character string spliced is pasted in the picture having a size of 1440x900 afterwards, and records character string position, by its according to
The width cutting of 16 pixels is character block for using when training.
Simultaneously because other than words identification, can also there is a large amount of Freehandhand-drawing geometry in the following Freehandhand-drawing blueprint for identification
Figure line, the ability for distinguishing text and geometric figure to have deep learning model in training increase dry in training set
Disturb line and interference text.Interfere lines can be by the image in elliptic curve, bezier curve, straight line and Freehandhand-drawing scribble training set
Random combine forms, as shown in Figure 8.
Finally, by generating magnanimity (such as 500,000) training set training pattern, so that model has the accurate text of identification
Block field mark function.
Optionally, second deep neural network is long including the second deep layer convolutional neural networks layer, the second bidirectional circulating
Short-term memory neural net layer, the second full articulamentum and the second output layer；
Above-mentioned steps 1023 include:
Using the label picture as the input of the second deep layer convolutional neural networks layer, pass through the second deep layer convolution mind
Feature extraction is carried out to label preselected area each in the label picture through network layer, obtains the second network characterization figure；
Using the second network characterization figure as the input of the long Memory Neural Networks layer in short-term of the second bidirectional circulating, pass through institute
It states the long Memory Neural Networks layer in short-term of the second bidirectional circulating and obtains third tensor；
Using the third tensor as the input of the second full articulamentum, the 4th is obtained by the described second full articulamentum
Amount；
Using the 4th tensor as the input of the second output layer, it is special that two network is obtained by second output layer
Every frame result vector is mapped as the probability of all icons and text in sign figure；
Be mapped as the probability of all icons and text according to frame result vector every in the two network characterizations figure, determine described in
The text for marking the icon for indicating landform element type in picture in each label preselected area and indicating landform element geometric parameter.
Here, Text region is predicted using end-to-end deep neural network, the design architecture and label preselected area of network
The first deep neural network it is similar, be equally convolutional neural networks+Recognition with Recurrent Neural Network framework, pass through convolutional neural networks
Extracting feature transfers to Recognition with Recurrent Neural Network to judge that final result, network using CTC-LOSS assessment of loss and optimize net in training
Network parameter.
Specific network parameter is as shown in the table, but not limited to this.
It is deep layer convolutional neural networks layer in the 1-14 layer of model, part pond layer uses rectangle form pool window in this process
Mouthful replace traditional rectangular pond window, make final characteristic pattern extraction result having a size of w/16 x 1, each 512 dimensional feature to
Amount represents the image information of receptive field (frame).
Since every frame not necessarily includes all information of letter, model can not be according to the feature vector independent judgment word of every frame
Symbol needs integrating context information, therefore all 512 dimensional vectors for representing every frame information is put into double-layer double-direction B-LSTM network,
Show that a 256 dimensional vector of final number of frames (W/16) is sent to final output layer.
Output layer can normalize exponential function by softmax and the result vector of every frame is mapped as all characters in dictionary
Probability, training when for calculate CTC-Loss (Connectionist temporal classification-Loss, connection
Timing Classification Loss) function backpropagation.
At this point, the final result exported by the second deep neural network, can recognize that in all label preselected areas
It indicates the icon of landform element type and indicates the text of landform element geometric parameter, so that it is determined that the type of landform element and several
What parameter.
Optionally, before above-mentioned steps 1023, further includes:
By pre-prepd second rough draft label data collection, the second deep neural network is trained, and in training
Increase interfering line and interference text in the process；Wherein, the second rough draft label data collection includes icon data set and text number
According to collection.
Here, similar with the deep neural network model of label preselected area, it is difficult to obtain what a large amount of real users were drawn
Icon text needs to construct training set using the mode that simulation generates, that is, passes through pre-prepd second rough draft label data
Collection is trained.Specifically, such as can be by being spliced into 500,000 surveys at random in the icon library and handwritten numeral training set collected in advance
Picture on probation is trained as the second rough draft label data collection, but not limited to this.While in order to enhance the generalization ability of model,
Random addition interfering line and interference text, as shown in Figure 9.
Optionally, above-mentioned steps 104 include:
Step 1041: going label picture to carry out an image thinning according to default thinning algorithm for described, and thin to image
Change that treated goes label picture to carry out expansion process and corrosion treatment, obtains first stage picture.
Step 1042: label picture will be gone to carry out logical AND behaviour before the first stage picture and image thinning processing
Make, obtains second stage picture.
Here, label picture profile being removed all relatively slightly after binaryzation, and there are some noises, we are rotten using expansion
Erosion, the mode of refinement are further processed.
Expansion is essentially identical with corrosion principle, the matrix that the value of a N*N is complete 1 is constructed, as a small window in full figure
Upper movement, moving step length is M (M≤N), when mobile every time, by the value and window square of the upper pixel being overlapped with window of figure
Battle array carries out with operation.In expansion process, if with operation result is all 0, the value of the pixel in window is all set
It is set to 0, is otherwise all set to 255.In corrosion process, if with operation is all 1, the value of the pixel in window is complete
Portion is set as 255, is otherwise all set to 0.
Simple first expansion, post-etching is as a result and bad, especially for it is some relatively close to lines, may weigh
A thick lines are synthesized, as shown in Figure 10.Therefore, we are using first carrying out image thinning, reflation, corrosion, then with two
Label picture is gone to do with operation after value, such result is just substantially consistent with original image, as shown in figure 11.Wherein thinning algorithm
Zhang-suen algorithm such as can be used, but not limited to this.
Step 1043: the second stage picture being subjected to secondary image refinement according to default thinning algorithm, obtains third
Stage picture.
Here, due to doing excessive erosion, expansion after refinement, profile still has many thick places, we reuse
Thinning algorithm (such as zhang-suen algorithm) refinement is primary, obtains the final image carried out before lines of outline identification that we want
Data, i.e. phase III picture, as shown in figure 12.
Step 1044: searching the profile of geometric figure contour line by way of seeking all in the phase III picture
Point, according to seek all over result determine in the phase III picture indicate landform element outline geometric figure.
At this point, the profile point of geometric figure contour line can be found by way of seeking all over, so that it is determined that going out to indicate ground
The geometric figure of shape element outline.
Search the embodiment of the present invention profile point of geometric figure contour line by way of seeking all over below one specifically answers
It is illustrated below with process.
Profile starting point is looked for first, is begun stepping through from the upper left corner, the point A that first pixel value is 255 is found, as first
The starting point of contour line judges 8 adjacent points of the point centered on the point.If only one point B value is that 255, A point is
One endpoint of the profile, B point are next points of profile；If there is 2 point B, C values are 255, then illustrate that A is not profile
Endpoint, need the direction along two points of B, C to continually look for the subsequent point of profile；It is 255 if it is greater than 2 point values, illustrates that A is
The crosspoint of multiple profiles will find suitable two points in multiple points, as the searching subsequent both direction of profile.It determines
The rule of two o'clock: the slope of these points and A point line is calculated, subsequent point of most similar two points of slope as the profile is taken.
As shown in figure 13, (0,7) (1,6) (2,5) (3,4) are best periphery two o'clock；
(0,4) (0,6) (1,5) (1,7) (2,3) (2,6) (3,7) (4,5) are next periphery two o'clock；
(0,5) (0,2) (1,3) (1,4) (2,7) (3,6) (4,6), (5,7) are periphery two o'clock again；
(0,1) (0,3) (1,2) (2,4) (3,5) (4,7) (5,6) (6,7) are worst periphery two o'clock.
0 is set by the ABC found (may be without C) value put, prevents from traversing the point again.
Then look for 8, the periphery point of B.If the value that the periphery of B is not put is 255, theoretically the contour line direction is answered
To this stopping.But actually Freehandhand-drawing error or the error after Morphological scale-space, it is possible to should be the point ined succession
It breaks, at this moment needs to carry out Error processing (breakpoint connecting).It is exactly to look for N number of point forward along the original direction of profile in principle
Range, if it is found, then think the point be subsequent point, otherwise contour line the direction terminate.As shown in figure 14, B point week
While not having adjacent value is 255 point, then the point of dash area is found, if there is being worth the point for being 255, then as the subsequent of B point
Point continually looks for, and then search of the profile in the direction B terminates if it is not found,.
If the value of only one point of the periphery of B is 255, necessarily next point of contour line.
If it is crosspoint that it is 255, B point that the periphery of B, which is greater than the value of a point, need to carry out selection judgement (crosspoint
Selection).It is equal to 255 peripheral point for each, as starting point, finds the son that all length is more than or equal to N (assuming that N=20)
Line segment seeks slope to the adjacent two o'clock of each sub-line section and seeks the slope average value of sliver, slope average value and former profile
Subsequent point of the most similar all the points as sub-line section of slope average value of last N number of point as profile.To choose sub-line section
The last one point continually look for 8, periphery point.
If there is C point, then according to the identical mode of 8, the periphery B point is found, the another part of profile is obtained.
It identifies in all contour lines, due to the presence of noise, has many short or small area useless wheels
Exterior feature, simultaneously because the presence of label, the contour line of label also belongs to useless contour line.In order to eliminate undesirable profile, optionally,
After step 1044, further includes:
The evaluation parameter for obtaining each geometric figure determined in the phase III picture, by the evaluation of each geometric figure
Parameter is compared with preset evaluation index respectively；
Delete the geometric figure that evaluation parameter in the phase III picture is less than preset evaluation index；
Wherein, the evaluation parameter includes profile minimum image vegetarian refreshments number and profile minimum circumscribed rectangle area.
Here, profile minimum image vegetarian refreshments number and profile minimum circumscribed rectangle area two indices are defined, by all wheels
Exterior feature is calculated by the two evaluation indexes, deletes the contour line for being less than index.Meanwhile identifying the minimum rectangle of outgoing label, it will
The contour line that label minimum rectangle includes is deleted.The interference of undesirable profile line is avoided, to accurately obtain expression landform member
The geometric figure of plain profile.
Optionally, above-mentioned steps 105 include:
Step 1051: the label that will identify that is opposite with each geometric figure according to target labels as target labels
Positional relationship determines the geometric figure to be matched of target labels；
Step 1052: if geometric figure to be matched, which does not match, gives other labels, using the geometric figure to be matched as mesh
Mark the matching geometric figure of label；
Step 1053: if geometric figure to be matched, which has matched, gives other labels, obtaining target labels minimum circumscribed rectangle
First ratio of intersection and union with geometric figure minimum circumscribed rectangle to be matched, and obtain and matched the minimum external square of label
The intersection of shape and geometric figure minimum circumscribed rectangle to be matched and the second ratio of union；
Step 1054: if the first ratio is greater than the second ratio, using geometric figure to be matched as the matching of target labels
Geometric figure；If the second ratio is greater than the first ratio, geometric figure to be matched is continued several as the matching for having matched label
What figure；
Step 1055: if the first ratio and the second ratio are identical, obtain target labels minimum circumscribed rectangle central point with
The first distance of geometric figure minimum circumscribed rectangle central point to be matched, and obtain and matched label minimum circumscribed rectangle central point
With the second distance of geometric figure minimum circumscribed rectangle central point to be matched；
Step 1056: if first distance is less than second distance, using geometric figure to be matched as the matching of target labels
Geometric figure；If second distance is greater than first distance, geometric figure to be matched is continued several as the matching for having matched label
What figure.
At this point, it is basic as evaluation index using the minimum circumscribed rectangle of geometric figure and the minimum circumscribed rectangle of label,
The geometric figure of tag match is accurately obtained.
Optionally, in step 1051, according to the relative positional relationship of target labels and each geometric figure, target mark is determined
Label geometric figure to be matched the step of include:
Searching overlay area and target labels has the geometric figure of intersection；
If the overlay area of only one geometric figure and the overlay area of target labels have intersection, by the geometric graph
To be matched geometric figure of the shape as target labels；
If having the overlay area of multiple geometric figures and the overlay area of target labels that there is intersection, there is friendship by multiple
The geometric figure of collection obtains target labels minimum circumscribed rectangle and each candidate figure minimum circumscribed rectangle as candidate figure
The ratio of intersection and union, and obtain to be matched geometric figure of the maximum candidate figure of ratio as target labels；
If being not covered with region and target labels has the geometric figure of intersection, target labels minimum circumscribed rectangle is obtained
Central point is obtained apart from the smallest geometric figure at a distance from each geometric figure minimum circumscribed rectangle central point as target
The geometric figure to be matched of label.
At this point, it is basic as evaluation index using the minimum circumscribed rectangle of geometric figure and the minimum circumscribed rectangle of label,
The geometric figure to be matched of label is obtained first, then the geometric figure of tag match is obtained by abovementioned steps 1052-1056.
Wherein, it includes that the overlay area of geometric figure includes mesh that the overlay area of geometric figure and target labels, which have intersection,
The overlay area of mark label and geometric figure intersects two kinds of situations with target labels.
Optionally, above-mentioned steps 106 include:
Step 1061: according to the position of the matched label of geometric figure, geometric figure and geometric figure that identify, respectively
Determine profile, type, geometric parameter and the position of rough draft topographic map mesorelief element.
Step 1062: according to the type and geometric parameter of landform element, obtaining the gray scale of performance landform element stereoscopic effect
Figure.
Here, up-stream system has had identified the position of landform element, chamfered shape and some main geometric parameters
Number, but these will realize real element or far from being enough, such as mountains and rivers ascending curve in map is not
May be simple straight line, in order to meet reality to have it is certain rise and fall, mountain ridge mountain valley etc., and geometric parameters can also be reached
Several requirements.Further requirement may need to realize the mountain of designated shape or even the mountain of a text.In order to reach these
Omnifarious demand final decision forms mountains and rivers by the way of grayscale image.
The advantage of grayscale image is that almost can all digitize all demands, is then stored in a picture.
This picture can be almost limitless according to size mountains and rivers precision.And it is easy to other people to check and show and final share to
Other users.It even can finally become the gray scale picture library of specified landform.
Step 1063: profile and grayscale image to landform element carry out distortion processing respectively, according to the landform member after distortion
Element profile and grayscale image obtain performance landform element stereoscopic effect, and with the matched height map of landform element shape.
Here, it is rectangle that a disadvantage of picture, which is shape, but the chamfered shape of the landform element obtained is because of Freehandhand-drawing
Or various other reasons, it is difficult well-regulated shape.But stiff the covering of grayscale image can not be cut out in designated shape
It cuts and will form the discontinuous of pixel in this way.There are also grayscale image generally have a large amount of blank, it is hollow situations such as, equally will lead to life
At landform can be less than specified landform (because empty side).So this step takes dual twisting technique, i.e., distort simultaneously
The profile of grayscale image and landform element reaches matching, carries out the production of height map later.
Specifically, as shown in figure 15, the curve in picture can be removed two sides X, Y when the profile of distortion landform element
Then to minimum and maximum two limiting values, go out minimum circumscribed rectangle.Parity method is penetrated by ray again and detects all songs
Point in all curves is mapped in minimum circumscribed rectangle region by the point in line finally by distortion formula.
It, as shown in figure 16, can mostly a kind of folded processing mode in the above case said for concave polygon.Namely work as knowledge
When point between other concave point two sides ab or cd, it is possible to determine that in the case where in polygon, twist into the radius of rectangle
The corresponding ab length of point of use (or cd length).It will form in rectangular area point in this way to correspond in polygon
Multiple the case where.
The case where purpose for distorting grayscale image is to remove empty side allows all gray scales full of this rectangular area.It can
To use plavini, is first progressively scanned in X-direction, obtain the aggregate-value of width.Then contracting is being formed according to the width of rectangle
The factor is put, the position for being only accumulated to rectangle originally is calculated by this zoom factor.Then it scans by column again, principle and X-axis
Direction operation is the same.Grayscale image is as shown in figure 17 after distortion.
It finally obtained two histograms, two histograms be scaled to length-width ratio is consistent, and positions all so are just formed
Then one-to-one effect generates height maps using two histograms.
Step 1064: according to the position of landform element, the height map of all landform elements being added to pre-generated three
In dimension word matrix.
Here, the determination landform function to be stored, first is to have specific performance, and high mountain, river and grassland will have bright
Aobvious difference, second is the three-dimensional performance power of three-dimensional space, and third is that landform easily supports later additions and deletions change to look into operation.So adopting
With three-dimensional character matrix is established, All Around The World is digitized, portion three-dimensional picture pasting both can be used in this way and clearly showed, had simultaneously
Standby three-dimensional sense, more because being the content that matrix can fast and accurately position designated precision position.
The height map of all landform elements is all added among the 3-dimensional digital matrix of generation, to be based on 3-dimensional digital
Matrix obtains electronic topographic map.Wherein, it should be noted that sequence when superposition, first solid addition such as can be used, then carved in hollow out
It carves.For example Plain mountains and rivers are first stacked, then cutting down rivers and lakes etc. out.Final 3-dimensional digital square has been eventually formed in this way
Battle array.
Step 1065: according to superimposed 3-dimensional digital matrix, generating editable electronic topographic map.
The rough draft topographic map of player can be converted into game editor automatically by the topographic map conversion method of the embodiment of the present invention
Editable electronics blueprint in device, player need to only be determined the geomorphic type in specified region by icon in rough draft topographic map, led to
Cross geometric figure lines and describe landforms chamfered shape, without executing other operations, implementation is simple, conveniently, it is easy to operate, and mention
High treatment effeciency.
In some embodiments of the invention, referring to Fig.1 shown in 8, a kind of topographic map converting system is additionally provided, comprising:
Preprocessing module 181, for pre-processing to rough draft topographic map, interference picture is removed in acquisition；
First identification module 182 goes in interference picture to indicate landform element described in identification for passing through deep neural network
The label of type and geometric parameter；
Removing module 183, the label for will identify that go in interference picture to delete from described, and label picture is removed in acquisition；
Second identification module 184, for identification geometric figure for going in label picture to indicate landform element outline；
Matching module 185, the label for will identify that are matched with the geometric figure identified；
Conversion module 186, for the position according to the matched label of geometric figure, geometric figure and geometric figure identified
It sets, rough draft topographic map is converted into editable electronic topographic map.
The rough draft topographic map of player can be converted into game editor automatically by the topographic map converting system of the embodiment of the present invention
Editable electronics blueprint in device, player need to only be determined the geomorphic type in specified region by icon in rough draft topographic map, led to
Cross geometric figure lines and describe landforms chamfered shape, without executing other operations, implementation is simple, conveniently, it is easy to operate, and mention
High treatment effeciency.
Optionally, the preprocessing module 181 includes:
First processing submodule obtains binaryzation picture for carrying out binary conversion treatment to rough draft topographic map；
Second processing submodule, for binaryzation picture image noise and interfering line be removed processing, gone
Interfere picture.
Optionally, first identification module 182 includes:
First identification submodule described removes table in interference picture for by the first deep neural network, identifying and intercepting
Show the label preselected area of landform element type and geometric parameter；
Submodule is removed, after expanding the range of label preselected area by parameter preset, in label preselected area
Interfering line be removed processing, and adjust the coordinate position of label preselected area, obtain the mark for being marked with the preparatory region of label
Remember picture；
Second identification submodule, for by the second deep neural network, identifying each label pre-selection in the label picture
The icon of landform element type is indicated in region and indicates the text of landform element geometric parameter.
Optionally, first deep neural network is long including the first deep layer convolutional neural networks layer, the first bidirectional circulating
Short-term memory neural net layer, the first full articulamentum and the first output layer；
The first identification submodule is specifically used for:
It goes interference picture as the input of the first deep layer convolutional neural networks layer for described, passes through the first deep layer convolution
Neural net layer goes interference picture to carry out feature extraction to described, obtains first network characteristic pattern；
The first network characteristic pattern is slided, every row obtains W with the window of predetermined size to slide unit by turn line by line
A one-dimensional vector, W are the width of the first network characteristic pattern；
Using the W one-dimensional vector that every row obtains as the defeated of the long Memory Neural Networks layer in short-term of the first bidirectional circulating
Enter, the first tensor is obtained by the long Memory Neural Networks layer in short-term of first bidirectional circulating；
Using first tensor as the input of the first full articulamentum, second is obtained by the described first full articulamentum
Amount；
Using second tensor as the input of the first output layer, the first network is obtained by first output layer
Each pixel mapped goes the output of pixel region in interference picture as a result, the output result includes vertical sits in characteristic pattern
Mark offset prediction result, text probabilistic forecasting result and boundary shifts amount prediction result；
According to pixel mapped each in the network characterization figure go interference picture in pixel region output as a result,
It goes in interference picture to indicate the label preselected area of landform element type and geometric parameter described in determination.
Optionally, the first identification submodule is also used to:
Using second tensor as the input of the first output layer, the first network is obtained by first output layer
Each pixel mapped under different anchor anchor sizes goes the output result of pixel region in interference picture in characteristic pattern.
Optionally, second deep neural network is long including the second deep layer convolutional neural networks layer, the second bidirectional circulating
Short-term memory neural net layer, the second full articulamentum and the second output layer；
The second identification submodule is specifically used for:
Using the label picture as the input of the second deep layer convolutional neural networks layer, pass through the second deep layer convolution mind
Feature extraction is carried out to label preselected area each in the label picture through network layer, obtains the second network characterization figure；
Using the second network characterization figure as the input of the long Memory Neural Networks layer in short-term of the second bidirectional circulating, pass through institute
It states the long Memory Neural Networks layer in short-term of the second bidirectional circulating and obtains third tensor；
Using the third tensor as the input of the second full articulamentum, the 4th is obtained by the described second full articulamentum
Amount；
Using the 4th tensor as the input of the second output layer, it is special that two network is obtained by second output layer
Every frame result vector is mapped as the probability of all icons and text in sign figure；
Be mapped as the probability of all icons and text according to frame result vector every in the two network characterizations figure, determine described in
The text for marking the icon for indicating landform element type in picture in each label preselected area and indicating landform element geometric parameter.
Optionally, system further include:
First training module, for passing through pre-prepd first rough draft label data collection, to the first deep neural network
It is trained, and increases interfering line and interference text in the training process；
Wherein, the first rough draft label data collection includes icon data set and lteral data collection；
Second training module, for passing through pre-prepd second rough draft label data collection, to the second deep neural network
It is trained, and increases interfering line and interference text in the training process；
Wherein, the second rough draft label data collection includes icon data set and lteral data collection.
Optionally, second identification module 184 includes:
Third handles submodule, for going label picture to carry out an image thinning according to default thinning algorithm for described,
And to image thinning, treated that label picture is gone to carry out expansion process and corrosion treatment, obtains first stage picture；
Logical AND operate submodule, for by the first stage picture and image thinning processing before go label picture into
The operation of row logical AND, obtains second stage picture；
Secondary refinement submodule is thin for the second stage picture to be carried out secondary image according to default thinning algorithm
Change, obtains phase III picture；
Submodule is sought all over, for searching geometric figure contour line by way of seeking all in the phase III picture
Profile point, according to seek all over result determine in the phase III picture indicate landform element outline geometric figure.
Optionally, system further include:
Comparison module will be each for obtaining the evaluation parameter for each geometric figure determined in the phase III picture
The evaluation parameter of geometric figure is compared with preset evaluation index respectively；
Removing module, the geometric graph for being less than preset evaluation index for deleting evaluation parameter in the phase III picture
Shape；
Wherein, the evaluation parameter includes profile minimum image vegetarian refreshments number and profile minimum circumscribed rectangle area.
Optionally, the matching module 185 includes:
First determines submodule, and the label for will identify that is as target labels, according to target labels and each geometry
The relative positional relationship of figure determines the geometric figure to be matched of target labels；
First matched sub-block gives other labels if not matching for geometric figure to be matched, by the geometry to be matched
Matching geometric figure of the figure as target labels；
First acquisition submodule gives other labels if having matched for geometric figure to be matched, obtains target labels most
The intersection of small boundary rectangle and geometric figure minimum circumscribed rectangle to be matched and the first ratio of union, and obtain and matched label
The intersection of minimum circumscribed rectangle and geometric figure minimum circumscribed rectangle to be matched and the second ratio of union；
Second matched sub-block, if being greater than the second ratio for the first ratio, using geometric figure to be matched as target
The matching geometric figure of label；If the second ratio be greater than the first ratio, by geometric figure to be matched continue as matched mark
The matching geometric figure of label；
Second acquisition submodule obtains the minimum external square of target labels if identical for the first ratio and the second ratio
The first distance of shape central point and geometric figure minimum circumscribed rectangle central point to be matched, and obtain that have matched label minimum external
The second distance of rectangular centre point and geometric figure minimum circumscribed rectangle central point to be matched；
Third matched sub-block, if being less than second distance for first distance, using geometric figure to be matched as target
The matching geometric figure of label；If second distance be greater than first distance, by geometric figure to be matched continue as matched mark
The matching geometric figure of label.
Optionally, described first determine that submodule includes:
Searching unit has the geometric figure of intersection for searching overlay area and target labels；
First determination unit, if having for the overlay area of only one geometric figure and the overlay area of target labels
Intersection, then using the geometric figure as the geometric figure to be matched of target labels；
Second determination unit, if being handed over for there is the overlay area of the overlay area of multiple geometric figures and target labels to have
Collection obtains target labels minimum circumscribed rectangle and each candidate then using multiple geometric figures with intersection as candidate figure
The intersection of figure minimum circumscribed rectangle and the ratio of union, and obtain ratio maximum candidate figure as target labels to
With geometric figure；
Third determination unit obtains mesh if having the geometric figure of intersection for being not covered with region and target labels
Label minimum circumscribed rectangle central point is marked at a distance from each geometric figure minimum circumscribed rectangle central point, and it is minimum to obtain distance
To be matched geometric figure of the geometric figure as target labels.
Optionally, the conversion module 186 includes:
Second determines submodule, for according to the matched label of geometric figure, geometric figure and geometric figure identified
Position, respectively determine rough draft topographic map mesorelief element profile, type, geometric parameter and position；
It is three-dimensional to obtain performance landform element for the type and geometric parameter according to landform element for third acquisition submodule
The grayscale image of effect；
4th acquisition submodule, for landform element profile and grayscale image carry out distortion processing respectively, according to distortion
The profile and grayscale image of landform element afterwards obtain performance landform element stereoscopic effect, and with the matched height of landform element shape
Figure；
It is superimposed submodule, for the position according to landform element, the height map of all landform elements is added to pre- Mr.
At 3-dimensional digital matrix in；
Submodule is generated, for generating editable electronic topographic map according to superimposed 3-dimensional digital matrix.
The rough draft topographic map of player can be converted into game editor automatically by the topographic map converting system of the embodiment of the present invention
Editable electronics blueprint in device, player need to only be determined the geomorphic type in specified region by icon in rough draft topographic map, led to
Cross geometric figure lines and describe landforms chamfered shape, without executing other operations, implementation is simple, conveniently, it is easy to operate, and mention
High treatment effeciency.
Topographic map converting system provided in an embodiment of the present invention is able to achieve each mistake in the embodiment of the method for Fig. 1 to Figure 17
Journey, and identical technical effect can be reached, to avoid repeating, details are not described herein.The topographic map of the embodiment of the present invention converts system
System,
The embodiment of the present invention also provides a kind of computer readable storage medium, and meter is stored on computer readable storage medium
Calculation machine program, the computer program realize each process of above-mentioned topographic map conversion method embodiment when being executed by processor, and
Identical technical effect can be reached, to avoid repeating, which is not described herein again.Wherein, the computer readable storage medium, such as
Read-only memory (Read-Only Memory, abbreviation ROM), random access memory (Random Access Memory, abbreviation
RAM), magnetic or disk etc..
It should be noted that, in this document, the terms "include", "comprise" or its any other variant are intended to non-row
His property includes, so that the process, method, article or the device that include a series of elements not only include those elements, and
And further include other elements that are not explicitly listed, or further include for this process, method, article or device institute it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including being somebody's turn to do
There is also other identical elements in the process, method of element, article or device.
Through the above description of the embodiments, those skilled in the art can be understood that above-described embodiment side
Method can be realized by means of software and necessary general hardware platform, naturally it is also possible to by hardware, but in many cases
The former is more preferably embodiment.Based on this understanding, technical solution of the present invention substantially in other words does the prior art
The part contributed out can be embodied in the form of software products, which is stored in a storage medium
In (such as ROM/RAM, magnetic disk, CD), including some instructions are used so that a terminal (can be mobile phone, computer, service
Device, air conditioner or network equipment etc.) execute method described in each embodiment of the present invention.
The embodiment of the present invention is described with above attached drawing, but the invention is not limited to above-mentioned specific
Embodiment, the above mentioned embodiment is only schematical, rather than restrictive, those skilled in the art
Under the inspiration of the present invention, without breaking away from the scope protected by the purposes and claims of the present invention, it can also make very much
Form belongs within protection of the invention.
Claims (10)
1. a kind of topographic map conversion method characterized by comprising
Rough draft topographic map is pre-processed, interference picture is removed in acquisition；
By deep neural network, go in interference picture to indicate the label of landform element type and geometric parameter described in identification；
The label that will identify that goes in interference picture to delete from described, and label picture is removed in acquisition；
It goes in label picture to indicate the geometric figure of landform element outline described in identification；
The label that will identify that is matched with the geometric figure identified；
According to the position of the matched label of geometric figure, geometric figure and geometric figure that identify, rough draft topographic map is converted
At editable electronic topographic map.
2. acquisition goes to do the method according to claim 1, wherein described pre-process rough draft topographic map
The step of disturbing picture include:
Binary conversion treatment is carried out to rough draft topographic map, obtains binaryzation picture；
Image noise and interfering line to binaryzation picture are removed processing, and interference picture is removed in acquisition.
3. going to interfere described in identification the method according to claim 1, wherein described by deep neural network
The step of label of expression landform element type and geometric parameter, includes: in picture
By the first deep neural network, identifies and intercept and described go in interference picture to indicate landform element type and geometric parameter
Label preselected area；
After the range of label preselected area is expanded by parameter preset, place is removed to the interfering line in label preselected area
Reason, and the coordinate position of label preselected area is adjusted, obtain the label picture for being marked with the preparatory region of label；
By the second deep neural network, identifying in the label picture indicates landform element type in each label preselected area
Icon and the text for indicating landform element geometric parameter.
4. according to the method described in claim 3, it is characterized in that, first deep neural network includes the first deep layer convolution
Neural net layer, the first bidirectional circulating long Memory Neural Networks layer, the first full articulamentum and the first output layer in short-term；
It is described by the first deep neural network, identify and intercept and described go in interference picture to indicate landform element type and geometry
The step of label preselected area of parameter includes:
It goes interference picture as the input of the first deep layer convolutional neural networks layer for described, passes through the first deep layer convolutional Neural
Network layer goes interference picture to carry out feature extraction to described, obtains first network characteristic pattern；
The first network characteristic pattern is slided, every row obtains W one with the window of predetermined size to slide unit by turn line by line
Dimensional vector, W are the width of the first network characteristic pattern；
Using the W one-dimensional vector that every row obtains as the input of the long Memory Neural Networks layer in short-term of the first bidirectional circulating, lead to
It crosses the long Memory Neural Networks layer in short-term of first bidirectional circulating and obtains the first tensor；
Using first tensor as the input of the first full articulamentum, the second tensor is obtained by the described first full articulamentum；
Using second tensor as the input of the first output layer, the first network feature is obtained by first output layer
Each pixel mapped goes the output of pixel region in interference picture as a result, the output result includes that ordinate is inclined in figure
Shifting amount prediction result, text probabilistic forecasting result and boundary shifts amount prediction result；
The output of pixel region in interference picture is gone according to pixel mapped each in the network characterization figure as a result, determining
The label preselected area for going in interference picture to indicate landform element type and geometric parameter.
5. the method according to claim 1, wherein going to indicate landform element in label picture described in the identification
The step of geometric figure of profile includes:
Label picture is gone to carry out an image thinning according to default thinning algorithm by described, and treated goes to mark to image thinning
It signs picture and carries out expansion process and corrosion treatment, obtain first stage picture；
Label picture will be gone to carry out logical AND operation before the first stage picture and image thinning processing, obtains second stage
Picture；
The second stage picture is subjected to secondary image refinement according to default thinning algorithm, obtains phase III picture；
The profile point for searching geometric figure contour line by way of seeking all in the phase III picture, according to seeking result all over
Determine the geometric figure that landform element outline is indicated in the phase III picture.
6. according to the method described in claim 5, it is characterized in that, being looked into the way of seeking all in the phase III picture
The profile point for looking for geometric figure contour line determines expression landform element outline in the phase III picture according to result is sought all over
After geometric figure, further includes:
The evaluation parameter for obtaining each geometric figure determined in the phase III picture, by the evaluation parameter of each geometric figure
It is compared respectively with preset evaluation index；
Delete the geometric figure that evaluation parameter in the phase III picture is less than preset evaluation index；
Wherein, the evaluation parameter includes profile minimum image vegetarian refreshments number and profile minimum circumscribed rectangle area.
7. the method according to claim 1, wherein the label that will identify that and the geometric figure identified
The step of being matched include:
The label that will identify that is determined as target labels according to the relative positional relationship of target labels and each geometric figure
The geometric figure to be matched of target labels；
It is if geometric figure to be matched, which does not match, gives other labels, the geometric figure to be matched is several as the matching of target labels
What figure；
If geometric figure to be matched, which has matched, gives other labels, target labels minimum circumscribed rectangle and geometric graph to be matched are obtained
The intersection of shape minimum circumscribed rectangle and the first ratio of union, and obtain and matched label minimum circumscribed rectangle and geometry to be matched
The intersection of figure minimum circumscribed rectangle and the second ratio of union；
If the first ratio is greater than the second ratio, using geometric figure to be matched as the matching geometric figure of target labels；If the
Two ratios are greater than the first ratio, then continue geometric figure to be matched as the matching geometric figure for having matched label；
If the first ratio and the second ratio are identical, target labels minimum circumscribed rectangle central point and geometric figure to be matched are obtained
The first distance of minimum circumscribed rectangle central point, and obtain and matched label minimum circumscribed rectangle central point and geometric graph to be matched
The second distance of shape minimum circumscribed rectangle central point；
If first distance is less than second distance, using geometric figure to be matched as the matching geometric figure of target labels；If the
Two distances are greater than first distance, then continue geometric figure to be matched as the matching geometric figure for having matched label.
8. the method according to the description of claim 7 is characterized in that described opposite with each geometric figure according to target labels
Positional relationship, the step of determining the geometric figure to be matched of target labels include:
Searching overlay area and target labels has the geometric figure of intersection；
If the overlay area of only one geometric figure and the overlay area of target labels have intersection, which is made
For the geometric figure to be matched of target labels；
If have the overlay area of multiple geometric figures and the overlay area of target labels that there is intersection, by multiple with intersection
Geometric figure obtains the intersection of target labels minimum circumscribed rectangle and each candidate figure minimum circumscribed rectangle as candidate figure
With the ratio of union, and to be matched geometric figure of the maximum candidate figure of ratio as target labels is obtained；
If being not covered with region and target labels has the geometric figure of intersection, target labels minimum circumscribed rectangle center is obtained
Point is obtained apart from the smallest geometric figure at a distance from each geometric figure minimum circumscribed rectangle central point as target labels
Geometric figure to be matched.
9. the method according to claim 1, wherein geometric figure, geometric figure that the basis identifies
The position of the label and geometric figure matched, the step of rough draft topographic map is converted into editable electronic topographic map include:
According to the position of the matched label of geometric figure, geometric figure and geometric figure that identify, rough draft landform is determined respectively
Profile, type, geometric parameter and the position of figure mesorelief element；
According to the type and geometric parameter of landform element, the grayscale image of performance landform element stereoscopic effect is obtained；
Profile and grayscale image to landform element carry out distortion processing respectively, according to the profile and gray scale of the landform element after distortion
Figure obtain performance landform element stereoscopic effect, and with the matched height map of landform element shape；
According to the position of landform element, in the 3-dimensional digital matrix for being added to pre-generated by the height map of all landform elements；
According to superimposed 3-dimensional digital matrix, editable electronic topographic map is generated.
10. a kind of topographic map converting system characterized by comprising
Preprocessing module, for pre-processing to rough draft topographic map, interference picture is removed in acquisition；
First identification module, for by deep neural network, go in interference picture to indicate described in identification landform element type and
The label of geometric parameter；
Removing module, the label for will identify that go in interference picture to delete from described, and label picture is removed in acquisition；
Second identification module, for identification geometric figure for going in label picture to indicate landform element outline；
Matching module, the label for will identify that are matched with the geometric figure identified；
Conversion module will be careless for the position according to the matched label of geometric figure, geometric figure and geometric figure identified
Original text topographic map is converted into editable electronic topographic map.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910453305.0A CN110180186B (en) | 2019-05-28 | 2019-05-28 | Topographic map conversion method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910453305.0A CN110180186B (en) | 2019-05-28 | 2019-05-28 | Topographic map conversion method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110180186A true CN110180186A (en) | 2019-08-30 |
| CN110180186B CN110180186B (en) | 2022-08-19 |
Family
ID=67718332
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910453305.0A Expired - Fee Related CN110180186B (en) | 2019-05-28 | 2019-05-28 | Topographic map conversion method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110180186B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110969659A (en) * | 2019-10-31 | 2020-04-07 | 浙江未来技术研究院(嘉兴) | Space positioning device and method for passive marking point |
| CN111199194A (en) * | 2019-12-25 | 2020-05-26 | 吉林大学 | Automobile intelligent cabin instrument testing method based on machine vision and deep learning |
| CN111744187A (en) * | 2020-08-10 | 2020-10-09 | 腾讯科技（深圳）有限公司 | Game data processing method and device, computer and readable storage medium |
| CN111870953A (en) * | 2020-07-24 | 2020-11-03 | 上海米哈游天命科技有限公司 | Height map generation method, device, equipment and storage medium |
| CN111957045A (en) * | 2020-09-01 | 2020-11-20 | 网易（杭州）网络有限公司 | Terrain deformation method, device, equipment and storage medium |
| CN112711965A (en) * | 2019-10-24 | 2021-04-27 | 深圳市优必选科技股份有限公司 | Method, device and equipment for identifying picture book |
| CN112862929A (en) * | 2021-03-10 | 2021-05-28 | 网易（杭州）网络有限公司 | Method, device and equipment for generating virtual target model and readable storage medium |
| CN113158982A (en) * | 2021-05-17 | 2021-07-23 | 广东中卡云计算有限公司 | Semi-intrusive target key point marking method |
| CN114088063A (en) * | 2021-10-19 | 2022-02-25 | 青海省交通工程技术服务中心 | Pier local scour terrain measurement method based on mobile terminal |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5111516A (en) * | 1989-04-11 | 1992-05-05 | Kabushiki Kaisha Toyota Chuo Kenkyusho | Apparatus for visual recognition |
| EP0495697A1 (en) * | 1991-01-15 | 1992-07-22 | Thomson-Csf | Image-shape recognition method |
| US7016536B1 (en) * | 1999-11-24 | 2006-03-21 | Gtx Corporation | Method and apparatus for automatic cleaning and enhancing of scanned documents |
| US20080220873A1 (en) * | 2007-03-06 | 2008-09-11 | Robert Ernest Lee | Distributed network architecture for introducing dynamic content into a synthetic environment |
| US20090275414A1 (en) * | 2007-03-06 | 2009-11-05 | Trion World Network, Inc. | Apparatus, method, and computer readable media to perform transactions in association with participants interacting in a synthetic environment |
| US20160104039A1 (en) * | 2014-10-10 | 2016-04-14 | Morpho | Method for identifying a sign on a deformed document |
| WO2019000653A1 (en) * | 2017-06-30 | 2019-01-03 | 清华大学深圳研究生院 | Image target identification method and apparatus |
| US20190122077A1 (en) * | 2016-03-15 | 2019-04-25 | Impra Europe S.A.S. | Method for classification of unique/rare cases by reinforcement learning in neural networks |
| KR101947650B1 (en) * | 2017-11-14 | 2019-05-20 | 국방과학연구소 | Apparatus and method for generating learning image in game engine-based machine learning |
-
2019
- 2019-05-28 CN CN201910453305.0A patent/CN110180186B/en not_active Expired - Fee Related
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5111516A (en) * | 1989-04-11 | 1992-05-05 | Kabushiki Kaisha Toyota Chuo Kenkyusho | Apparatus for visual recognition |
| EP0495697A1 (en) * | 1991-01-15 | 1992-07-22 | Thomson-Csf | Image-shape recognition method |
| US7016536B1 (en) * | 1999-11-24 | 2006-03-21 | Gtx Corporation | Method and apparatus for automatic cleaning and enhancing of scanned documents |
| US20080220873A1 (en) * | 2007-03-06 | 2008-09-11 | Robert Ernest Lee | Distributed network architecture for introducing dynamic content into a synthetic environment |
| US20090275414A1 (en) * | 2007-03-06 | 2009-11-05 | Trion World Network, Inc. | Apparatus, method, and computer readable media to perform transactions in association with participants interacting in a synthetic environment |
| US20160104039A1 (en) * | 2014-10-10 | 2016-04-14 | Morpho | Method for identifying a sign on a deformed document |
| US20190122077A1 (en) * | 2016-03-15 | 2019-04-25 | Impra Europe S.A.S. | Method for classification of unique/rare cases by reinforcement learning in neural networks |
| WO2019000653A1 (en) * | 2017-06-30 | 2019-01-03 | 清华大学深圳研究生院 | Image target identification method and apparatus |
| KR101947650B1 (en) * | 2017-11-14 | 2019-05-20 | 국방과학연구소 | Apparatus and method for generating learning image in game engine-based machine learning |
Non-Patent Citations (1)
| Title |
|---|
| 周蓓蓓: ""彩色地形图点状地物符号的提取与识别"", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑 计算机软件及计算机应用》, no. 11, 15 November 2008 (2008-11-15), pages 7 - 12 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112711965A (en) * | 2019-10-24 | 2021-04-27 | 深圳市优必选科技股份有限公司 | Method, device and equipment for identifying picture book |
| CN112711965B (en) * | 2019-10-24 | 2023-12-05 | 深圳市优必选科技股份有限公司 | Drawing recognition method, device and equipment |
| CN110969659B (en) * | 2019-10-31 | 2024-03-15 | 浙江清华长三角研究院 | Spatial positioning device and method for passive mark point |
| CN110969659A (en) * | 2019-10-31 | 2020-04-07 | 浙江未来技术研究院(嘉兴) | Space positioning device and method for passive marking point |
| CN111199194A (en) * | 2019-12-25 | 2020-05-26 | 吉林大学 | Automobile intelligent cabin instrument testing method based on machine vision and deep learning |
| CN111870953A (en) * | 2020-07-24 | 2020-11-03 | 上海米哈游天命科技有限公司 | Height map generation method, device, equipment and storage medium |
| CN111744187A (en) * | 2020-08-10 | 2020-10-09 | 腾讯科技（深圳）有限公司 | Game data processing method and device, computer and readable storage medium |
| CN111957045A (en) * | 2020-09-01 | 2020-11-20 | 网易（杭州）网络有限公司 | Terrain deformation method, device, equipment and storage medium |
| CN112862929A (en) * | 2021-03-10 | 2021-05-28 | 网易（杭州）网络有限公司 | Method, device and equipment for generating virtual target model and readable storage medium |
| CN112862929B (en) * | 2021-03-10 | 2024-05-28 | 网易（杭州）网络有限公司 | Method, device, equipment and readable storage medium for generating virtual target model |
| CN113158982A (en) * | 2021-05-17 | 2021-07-23 | 广东中卡云计算有限公司 | Semi-intrusive target key point marking method |
| CN114088063A (en) * | 2021-10-19 | 2022-02-25 | 青海省交通工程技术服务中心 | Pier local scour terrain measurement method based on mobile terminal |
| CN114088063B (en) * | 2021-10-19 | 2024-02-02 | 青海省交通工程技术服务中心 | Pier local scour terrain measurement method based on mobile terminal |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110180186B (en) | 2022-08-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110180186A (en) | A kind of topographic map conversion method and system | |
| CN109299274B (en) | Natural scene text detection method based on full convolution neural network | |
| CN110738207A (en) | character detection method for fusing character area edge information in character image | |
| CN110853026B (en) | Remote sensing image change detection method integrating deep learning and region segmentation | |
| CN113673338B (en) | Automatic labeling method, system and medium for weak supervision of natural scene text image character pixels | |
| Kadam et al. | Detection and localization of multiple image splicing using MobileNet V1 | |
| CN111723585A (en) | Style-controllable image text real-time translation and conversion method | |
| CN110287952B (en) | Method and system for recognizing characters of dimension picture | |
| CN111738055B (en) | Multi-category text detection system and bill form detection method based on same | |
| CN108345850A (en) | The scene text detection method of the territorial classification of stroke feature transformation and deep learning based on super-pixel | |
| US6917708B2 (en) | Handwriting recognition by word separation into silhouette bar codes and other feature extraction | |
| CN111553837A (en) | Artistic text image generation method based on neural style migration | |
| CN109948533B (en) | Text detection method, device and equipment and readable storage medium | |
| CN111860525B (en) | Bottom-up optical character recognition method suitable for terminal block | |
| CN111753828A (en) | Natural scene horizontal character detection method based on deep convolutional neural network | |
| CN112200117A (en) | Form identification method and device | |
| CN111950389B (en) | Depth binary feature facial expression recognition method based on lightweight network | |
| CN114937278A (en) | Text content extraction and identification method based on line text box word segmentation algorithm | |
| CN110517270A (en) | A kind of indoor scene semantic segmentation method based on super-pixel depth network | |
| CN109460767A (en) | Rule-based convex print bank card number segmentation and recognition methods | |
| CN112686265A (en) | Hierarchic contour extraction-based pictograph segmentation method | |
| CN109117841B (en) | Scene text detection method based on stroke width transformation and convolutional neural network | |
| CN112132750A (en) | Video processing method and device | |
| CN112800259B (en) | Image generation method and system based on edge closure and commonality detection | |
| CN113688864B (en) | Human-object interaction relation classification method based on split attention |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Yan Liting Inventor after: Zhang Ran Inventor before: Zhang Ran |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20220819 |